How much do we know about Stroud, Young and the rest of 2023 quarterback class?

We can't write-off or declare stardom for any quarterback after one season, but their results do tell us a lot about how to adjust our expectations

As the top-2 picks of the 2023 NFL draft, Bryce Young and C.J. Stroud will be forever intertwined, with the former’s team giving up a King’s ransom to trade up and select him over the latter, and both quarterbacks going through their final college seasons and the pre-draft buildup as the strongest candidates to be the best quarterback in the class. 

Nine months after the draft, we have a lot of evidence to make sense of for both quarterbacks. Both were their team’s Week 1 starter and played the vast majority of starter snaps for their teams, but their post-rookie impressions couldn’t be more different. We didn’t get the biggest sample of play from Anthony Richardson or Will Levis, but I’ll still discuss them in detail below.

Young’s rookie year was a disaster according to advanced stats and team results. The Panthers finished with a 2-15 record, and their new head coach didn’t even survive until December. Young accumulated -115 EPA over 620 dropbacks and 36 designed runs, giving him the second-worst efficiency of any quarterback with at least 400 dropbacks. Young’s 52.6 PFF passing grade was dead-last among the same group of qualifiers. 

On the other side of the spectrum, Stroud wowed stat- and film-watchers with his rookie year. Stroud finished 11th in EPA per play, better than Matthew Stafford, Trevor Lawrence and Justin Herbert, and his 79.4 PFF passing grade only slightly trailed that of Patrick Mahomes.

ASSESSING THE HISTORICAL COMPS

Looking at the larger landscape of rookie quarterbacks since PFF started charting in 2006, you can see that Stroud is in elite company, while Young certainly is not.

We shouldn’t measure Stroud for his Hall-of-Fame jacket yet, as a handful of quarterbacks with similar rookie seasons (e.g Robert Griffin and Baker Mayfield) have fooled fan bases into thinking they had found “the guy” in the past, but it’s clear the correlation of rookie-year and career-long success. For Young, you can find a couple examples of elite or serviceable quarterbacks who struggled as rookies, but not nearly as many as among those who thrived.

Of the 19 quarterbacks who had a sample of at least 400 plays as rookies and finished with above-average grading and efficiency, my subjective opinion would classify eight as strong hits (Russell Wilson, Dak Prescott, Matt Ryan, Justin Herbert, Joe Burrow, Cam Newton, Andrew Luck and Josh Allen), and another five as capable of being solid starters for several seasons (Baker Mayfield, Carson Wentz, Ryan Tannehill, Joe Flacco and Jameis Winston). That leaves only six outside of either bucket, with the careers for Kenny Pickett and Mac Jones only in their early stages.

Of the 16 quarterbacks who finished with below-average grading and efficiency, I’d only classify Matthew Stafford as a big hit, with Derek Carr becoming a serviceable starter. Trevor Lawrence was not good as a rookie, and three years into his career he looks likely to either finish as a hit or serviceable starter. For the quarterback with split grading and efficiency versus the averages, it looks like a feather in the cap of efficiency’s (EPA) cap for predictive power, as I’d much prefer to have the good-efficiency, poor-grade quarterbacks of Kyler Murray, Marcus Mariota and Andy Dalton over the inverse group of Daniel Jones, Sam Bradford and Mitchell Trubisky.

A BETTER WAY TO PROJECT QUARTERBACKS

The above analysis is convincing to the large degree that rookie-year performance matters for the future, but it also shows that we can’t rule out C.J. Stroud failing, or Bryce Young going on to an elite. The best way to resolve the uncertainties is to look at historical evidence and use historical tools to help us concretely quantify how likely it is that Stroud and Young will reach different thresholds of value.

Luckily I can lean, once again, on Bayesian updating to project exact ranges of outcomes for these and other rookies, using historical averages for the range of franchise starter efficiencies, and per-play averages and distributions. Each piece of evidence we get from players can be weighed against the normalized metrics, and we shift the range of outcomes while becoming more confident with more we learn. For Stroud and Young, we have relatively large rookie samples (600 and 664 play involvements, respectively), which will give us more narrow ranges of outcomes than for Anthony Richardson and Will Levis, who I’ll discuss more later.

Without getting into the minutiae, I’m using Bayesian updating with the a prior of the normal distribution. Historical franchise quarterback performances over their career and on a play-by-play basis match the normal distribution fairly well. We begin with the assumption that a quarterback will end up somewhere on the historical franchise-quarterback spectrum for efficiency, assuming they get enough reps to prove themselves. I don’t think it will be a problem for Stroud or Young, as both will have more than a thousand dropbacks sometime next season.

We then update our distribution, which encompasses the range of likelihoods for all possible outcomes, with each piece of evidence, in this case the EPA generated on each play involvement. As we gain evidence, a player’s projected range of outcomes will narrow and shift. The narrowness is based on the updated standard deviation, the shift on the updated mean.

After a year’s worth of adding evidence to our prior, we can project and visualize the distributions for Stroud and Young, with markers added for the level of efficiency for different franchise-quarterback percentiles, based on historical career numbers.